When Does Cryptographic Reference Integrity Add Value? Threat Conditions in the Verification of AI Recovery

RT5 asks when protected reference records improve decisions about AI recovery. It separates record availability, authenticity, freshness, measurement validity, and decision cost, and shows why failure depth alone cannot determine cryptographic value. A reproducible audit compares six decision policies across 24 fixed record fixtures. Signatures with an independently retained checkpoint expose several alterations, while withheld evidence can require abstention and authentic but initially false measurements can still mislead. These are protocol and counterexample checks rather than measurements of AI recovery. The contribution is a threat-conditioned assessment design; deployment benefit and universal depth-based prescriptions are not established.